CAREER: Efficient Predictive Modeling for Infrastructure Systems Using Polynomial Approximation

University Of Illinois At Urbana-Champaign, Urbana IL

Investigators

Abstract

Healthy and optimal operation of infrastructure systems will advance the health, prosperity and welfare of the society. It is however challenging to maintain the optimal condition given various uncertainties involved in these systems. Advances in computing and sensing technologies can potentially enable a paradigm shift in the management of infrastructure systems by realistically capturing the uncertainties involved. The main challenge is still the high computational cost that would entail, mainly due to large size of infrastructure networks. This Faculty Early Career Development Program (CAREER) award aims to enable the next generation of fast uncertainty quantification (UQ) methodologies that can significantly reduce simulation time and are particularly tailored for large infrastructure networks. This award will analyze the case of interdependent transportation-energy systems that involves the integration of electric vehicles. The project will also establish an integrated education and outreach plan to prepare the next generation of civil engineers with improved programming and computational skills. This is done through development of a new graduate-level course, mentoring of undergraduate researchers, partnership with educational physiologists to enhance educational plans, and also outreach to the broader civil engineering student population, K-12 students, and decision makers. The new methods will promote progress of science in UQ and infrastructure modeling, and together with the educational and outreach activities can collectively lead to improved infrastructure operations and promote the economic competitiveness of our society. This project will use stochastic simulations to realistically capture the complexities. The approaches will build upon and enrich current UQ machinery. These advanced UQ methods will aim to (1) intelligently identify the redundancies in the flow models of infrastructure systems, (2) use topology characteristics of these networks towards model reduction, (3) build an effective online training framework that use streaming sensor data, and (4) take advantage of the availability of flow models at various fidelity levels to produce multifidelity predictions. Specifically, the approaches include advanced compressive sampling methods in polynomial chaos expansion for effective removal of uncertainties related to network redundancies; a topology-based dimension reduction algorithm integrated with compressive sampling approach to exploit the topology information of infrastructure networks; a block-wise recursive least square approach to enable an effective online learning for polynomial surrogates with quantified modeling errors; and a multifidelity regression framework for building surrogates using results of simulations at various fidelity levels. If successful, these advances will collectively contribute to the transformation of infrastructure engineering where simulation-based data-intensive design suites that can assist decision makers will be increasingly used. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.